Upload struct_data_operators.py with huggingface_hub
Browse files- struct_data_operators.py +364 -0
struct_data_operators.py
ADDED
@@ -0,0 +1,364 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""This section describes unitxt operators for tabular data.
|
2 |
+
|
3 |
+
These operators are specialized in handling tabular data.
|
4 |
+
Input table format is assumed as:
|
5 |
+
{
|
6 |
+
"header": ["col1", "col2"],
|
7 |
+
"rows": [["row11", "row12"], ["row21", "row22"], ["row31", "row32"]]
|
8 |
+
}
|
9 |
+
|
10 |
+
------------------------
|
11 |
+
"""
|
12 |
+
import random
|
13 |
+
from abc import ABC, abstractmethod
|
14 |
+
from copy import deepcopy
|
15 |
+
from typing import (
|
16 |
+
Any,
|
17 |
+
Dict,
|
18 |
+
List,
|
19 |
+
Optional,
|
20 |
+
)
|
21 |
+
|
22 |
+
from .dict_utils import dict_get
|
23 |
+
from .operators import FieldOperator, StreamInstanceOperator
|
24 |
+
|
25 |
+
|
26 |
+
class SerializeTable(ABC, FieldOperator):
|
27 |
+
"""TableSerializer converts a given table into a flat sequence with special symbols.
|
28 |
+
|
29 |
+
Output format varies depending on the chosen serializer. This abstract class defines structure of a typical table serializer that any concrete implementation should follow.
|
30 |
+
"""
|
31 |
+
|
32 |
+
# main method to serialize a table
|
33 |
+
@abstractmethod
|
34 |
+
def serialize_table(self, table_content: Dict) -> str:
|
35 |
+
pass
|
36 |
+
|
37 |
+
# method to process table header
|
38 |
+
@abstractmethod
|
39 |
+
def process_header(self, header: List):
|
40 |
+
pass
|
41 |
+
|
42 |
+
# method to process a table row
|
43 |
+
@abstractmethod
|
44 |
+
def process_row(self, row: List, row_index: int):
|
45 |
+
pass
|
46 |
+
|
47 |
+
|
48 |
+
# Concrete classes implementing table serializers
|
49 |
+
class SerializeTableAsIndexedRowMajor(SerializeTable):
|
50 |
+
"""Indexed Row Major Table Serializer.
|
51 |
+
|
52 |
+
Commonly used row major serialization format.
|
53 |
+
Format: col : col1 | col2 | col 3 row 1 : val1 | val2 | val3 | val4 row 2 : val1 | ...
|
54 |
+
"""
|
55 |
+
|
56 |
+
def process_value(self, table: Any) -> Any:
|
57 |
+
table_input = deepcopy(table)
|
58 |
+
return self.serialize_table(table_content=table_input)
|
59 |
+
|
60 |
+
# main method that processes a table
|
61 |
+
# table_content must be in the presribed input format
|
62 |
+
def serialize_table(self, table_content: Dict) -> str:
|
63 |
+
# Extract headers and rows from the dictionary
|
64 |
+
header = table_content.get("header", [])
|
65 |
+
rows = table_content.get("rows", [])
|
66 |
+
|
67 |
+
assert header and rows, "Incorrect input table format"
|
68 |
+
|
69 |
+
# Process table header first
|
70 |
+
serialized_tbl_str = self.process_header(header) + " "
|
71 |
+
|
72 |
+
# Process rows sequentially starting from row 1
|
73 |
+
for i, row in enumerate(rows, start=1):
|
74 |
+
serialized_tbl_str += self.process_row(row, row_index=i) + " "
|
75 |
+
|
76 |
+
# return serialized table as a string
|
77 |
+
return serialized_tbl_str.strip()
|
78 |
+
|
79 |
+
# serialize header into a string containing the list of column names separated by '|' symbol
|
80 |
+
def process_header(self, header: List):
|
81 |
+
return "col : " + " | ".join(header)
|
82 |
+
|
83 |
+
# serialize a table row into a string containing the list of cell values separated by '|'
|
84 |
+
def process_row(self, row: List, row_index: int):
|
85 |
+
serialized_row_str = ""
|
86 |
+
row_cell_values = [
|
87 |
+
str(value) if isinstance(value, (int, float)) else value for value in row
|
88 |
+
]
|
89 |
+
|
90 |
+
serialized_row_str += " | ".join(row_cell_values)
|
91 |
+
|
92 |
+
return f"row {row_index} : {serialized_row_str}"
|
93 |
+
|
94 |
+
|
95 |
+
class SerializeTableAsMarkdown(SerializeTable):
|
96 |
+
"""Markdown Table Serializer.
|
97 |
+
|
98 |
+
Markdown table format is used in GitHub code primarily.
|
99 |
+
Format:
|
100 |
+
|col1|col2|col3|
|
101 |
+
|---|---|---|
|
102 |
+
|A|4|1|
|
103 |
+
|I|2|1|
|
104 |
+
...
|
105 |
+
"""
|
106 |
+
|
107 |
+
def process_value(self, table: Any) -> Any:
|
108 |
+
table_input = deepcopy(table)
|
109 |
+
return self.serialize_table(table_content=table_input)
|
110 |
+
|
111 |
+
# main method that serializes a table.
|
112 |
+
# table_content must be in the presribed input format.
|
113 |
+
def serialize_table(self, table_content: Dict) -> str:
|
114 |
+
# Extract headers and rows from the dictionary
|
115 |
+
header = table_content.get("header", [])
|
116 |
+
rows = table_content.get("rows", [])
|
117 |
+
|
118 |
+
assert header and rows, "Incorrect input table format"
|
119 |
+
|
120 |
+
# Process table header first
|
121 |
+
serialized_tbl_str = self.process_header(header)
|
122 |
+
|
123 |
+
# Process rows sequentially starting from row 1
|
124 |
+
for i, row in enumerate(rows, start=1):
|
125 |
+
serialized_tbl_str += self.process_row(row, row_index=i)
|
126 |
+
|
127 |
+
# return serialized table as a string
|
128 |
+
return serialized_tbl_str.strip()
|
129 |
+
|
130 |
+
# serialize header into a string containing the list of column names
|
131 |
+
def process_header(self, header: List):
|
132 |
+
header_str = "|{}|\n".format("|".join(header))
|
133 |
+
header_str += "|{}|\n".format("|".join(["---"] * len(header)))
|
134 |
+
return header_str
|
135 |
+
|
136 |
+
# serialize a table row into a string containing the list of cell values
|
137 |
+
def process_row(self, row: List, row_index: int):
|
138 |
+
row_str = ""
|
139 |
+
row_str += "|{}|\n".format("|".join(str(cell) for cell in row))
|
140 |
+
return row_str
|
141 |
+
|
142 |
+
|
143 |
+
# truncate cell value to maximum allowed length
|
144 |
+
def truncate_cell(cell_value, max_len):
|
145 |
+
if cell_value is None:
|
146 |
+
return None
|
147 |
+
|
148 |
+
if isinstance(cell_value, int) or isinstance(cell_value, float):
|
149 |
+
return None
|
150 |
+
|
151 |
+
if cell_value.strip() == "":
|
152 |
+
return None
|
153 |
+
|
154 |
+
if len(cell_value) > max_len:
|
155 |
+
return cell_value[:max_len]
|
156 |
+
|
157 |
+
return None
|
158 |
+
|
159 |
+
|
160 |
+
class TruncateTableCells(StreamInstanceOperator):
|
161 |
+
"""Limit the maximum length of cell values in a table to reduce the overall length.
|
162 |
+
|
163 |
+
Args:
|
164 |
+
max_length (int) - maximum allowed length of cell values
|
165 |
+
For tasks that produce a cell value as answer, truncating a cell value should be replicated
|
166 |
+
with truncating the corresponding answer as well. This has been addressed in the implementation.
|
167 |
+
|
168 |
+
"""
|
169 |
+
|
170 |
+
max_length: int = 15
|
171 |
+
table: str = None
|
172 |
+
text_output: Optional[str] = None
|
173 |
+
use_query: bool = False
|
174 |
+
|
175 |
+
def process(
|
176 |
+
self, instance: Dict[str, Any], stream_name: Optional[str] = None
|
177 |
+
) -> Dict[str, Any]:
|
178 |
+
table = dict_get(instance, self.table, use_dpath=self.use_query)
|
179 |
+
|
180 |
+
answers = []
|
181 |
+
if self.text_output is not None:
|
182 |
+
answers = dict_get(instance, self.text_output, use_dpath=self.use_query)
|
183 |
+
|
184 |
+
self.truncate_table(table_content=table, answers=answers)
|
185 |
+
|
186 |
+
return instance
|
187 |
+
|
188 |
+
# truncate table cells
|
189 |
+
def truncate_table(self, table_content: Dict, answers: Optional[List]):
|
190 |
+
cell_mapping = {}
|
191 |
+
|
192 |
+
# One row at a time
|
193 |
+
for row in table_content.get("rows", []):
|
194 |
+
for i, cell in enumerate(row):
|
195 |
+
truncated_cell = truncate_cell(cell, self.max_length)
|
196 |
+
if truncated_cell is not None:
|
197 |
+
cell_mapping[cell] = truncated_cell
|
198 |
+
row[i] = truncated_cell
|
199 |
+
|
200 |
+
# Update values in answer list to truncated values
|
201 |
+
if answers is not None:
|
202 |
+
for i, case in enumerate(answers):
|
203 |
+
answers[i] = cell_mapping.get(case, case)
|
204 |
+
|
205 |
+
|
206 |
+
class TruncateTableRows(FieldOperator):
|
207 |
+
"""Limits table rows to specified limit by removing excess rows via random selection.
|
208 |
+
|
209 |
+
Args:
|
210 |
+
rows_to_keep (int) - number of rows to keep.
|
211 |
+
"""
|
212 |
+
|
213 |
+
rows_to_keep: int = 10
|
214 |
+
|
215 |
+
def process_value(self, table: Any) -> Any:
|
216 |
+
return self.truncate_table_rows(table_content=table)
|
217 |
+
|
218 |
+
def truncate_table_rows(self, table_content: Dict):
|
219 |
+
# Get rows from table
|
220 |
+
rows = table_content.get("rows", [])
|
221 |
+
|
222 |
+
num_rows = len(rows)
|
223 |
+
|
224 |
+
# if number of rows are anyway lesser, return.
|
225 |
+
if num_rows <= self.rows_to_keep:
|
226 |
+
return table_content
|
227 |
+
|
228 |
+
# calculate number of rows to delete, delete them
|
229 |
+
rows_to_delete = num_rows - self.rows_to_keep
|
230 |
+
|
231 |
+
# Randomly select rows to be deleted
|
232 |
+
deleted_rows_indices = random.sample(range(len(rows)), rows_to_delete)
|
233 |
+
|
234 |
+
remaining_rows = [
|
235 |
+
row for i, row in enumerate(rows) if i not in deleted_rows_indices
|
236 |
+
]
|
237 |
+
table_content["rows"] = remaining_rows
|
238 |
+
|
239 |
+
return table_content
|
240 |
+
|
241 |
+
|
242 |
+
class SerializeTableRowAsText(StreamInstanceOperator):
|
243 |
+
"""Serializes a table row as text.
|
244 |
+
|
245 |
+
Args:
|
246 |
+
fields (str) - list of fields to be included in serialization.
|
247 |
+
to_field (str) - serialized text field name.
|
248 |
+
max_cell_length (int) - limits cell length to be considered, optional.
|
249 |
+
"""
|
250 |
+
|
251 |
+
fields: str
|
252 |
+
to_field: str
|
253 |
+
max_cell_length: Optional[int] = None
|
254 |
+
|
255 |
+
def process(
|
256 |
+
self, instance: Dict[str, Any], stream_name: Optional[str] = None
|
257 |
+
) -> Dict[str, Any]:
|
258 |
+
linearized_str = ""
|
259 |
+
for field in self.fields:
|
260 |
+
value = dict_get(instance, field, use_dpath=False)
|
261 |
+
if self.max_cell_length is not None:
|
262 |
+
truncated_value = truncate_cell(value, self.max_cell_length)
|
263 |
+
if truncated_value is not None:
|
264 |
+
value = truncated_value
|
265 |
+
|
266 |
+
linearized_str = linearized_str + field + " is " + str(value) + ", "
|
267 |
+
|
268 |
+
instance[self.to_field] = linearized_str
|
269 |
+
return instance
|
270 |
+
|
271 |
+
|
272 |
+
class SerializeTableRowAsList(StreamInstanceOperator):
|
273 |
+
"""Serializes a table row as list.
|
274 |
+
|
275 |
+
Args:
|
276 |
+
fields (str) - list of fields to be included in serialization.
|
277 |
+
to_field (str) - serialized text field name.
|
278 |
+
max_cell_length (int) - limits cell length to be considered, optional.
|
279 |
+
"""
|
280 |
+
|
281 |
+
fields: str
|
282 |
+
to_field: str
|
283 |
+
max_cell_length: Optional[int] = None
|
284 |
+
|
285 |
+
def process(
|
286 |
+
self, instance: Dict[str, Any], stream_name: Optional[str] = None
|
287 |
+
) -> Dict[str, Any]:
|
288 |
+
linearized_str = ""
|
289 |
+
for field in self.fields:
|
290 |
+
value = dict_get(instance, field, use_dpath=False)
|
291 |
+
if self.max_cell_length is not None:
|
292 |
+
truncated_value = truncate_cell(value, self.max_cell_length)
|
293 |
+
if truncated_value is not None:
|
294 |
+
value = truncated_value
|
295 |
+
|
296 |
+
linearized_str = linearized_str + field + ": " + str(value) + ", "
|
297 |
+
|
298 |
+
instance[self.to_field] = linearized_str
|
299 |
+
return instance
|
300 |
+
|
301 |
+
|
302 |
+
class SerializeTriples(FieldOperator):
|
303 |
+
"""Serializes triples into a flat sequence.
|
304 |
+
|
305 |
+
Sample input in expected format:
|
306 |
+
[[ "First Clearing", "LOCATION", "On NYS 52 1 Mi. Youngsville" ], [ "On NYS 52 1 Mi. Youngsville", "CITY_OR_TOWN", "Callicoon, New York"]]
|
307 |
+
|
308 |
+
Sample output:
|
309 |
+
First Clearing : LOCATION : On NYS 52 1 Mi. Youngsville | On NYS 52 1 Mi. Youngsville : CITY_OR_TOWN : Callicoon, New York
|
310 |
+
|
311 |
+
"""
|
312 |
+
|
313 |
+
def process_value(self, tripleset: Any) -> Any:
|
314 |
+
return self.serialize_triples(tripleset)
|
315 |
+
|
316 |
+
def serialize_triples(self, tripleset) -> str:
|
317 |
+
return " | ".join(
|
318 |
+
f"{subj} : {rel.lower()} : {obj}" for subj, rel, obj in tripleset
|
319 |
+
)
|
320 |
+
|
321 |
+
|
322 |
+
class SerializeKeyValPairs(FieldOperator):
|
323 |
+
"""Serializes key, value pairs into a flat sequence.
|
324 |
+
|
325 |
+
Sample input in expected format: {"name": "Alex", "age": 31, "sex": "M"}
|
326 |
+
Sample output: name is Alex, age is 31, sex is M
|
327 |
+
"""
|
328 |
+
|
329 |
+
def process_value(self, kvpairs: Any) -> Any:
|
330 |
+
return self.serialize_kvpairs(kvpairs)
|
331 |
+
|
332 |
+
def serialize_kvpairs(self, kvpairs) -> str:
|
333 |
+
serialized_str = ""
|
334 |
+
for key, value in kvpairs.items():
|
335 |
+
serialized_str += f"{key} is {value}, "
|
336 |
+
|
337 |
+
# Remove the trailing comma and space then return
|
338 |
+
return serialized_str[:-2]
|
339 |
+
|
340 |
+
|
341 |
+
class ListToKeyValPairs(StreamInstanceOperator):
|
342 |
+
"""Maps list of keys and values into key:value pairs.
|
343 |
+
|
344 |
+
Sample input in expected format: {"keys": ["name", "age", "sex"], "values": ["Alex", 31, "M"]}
|
345 |
+
Sample output: {"name": "Alex", "age": 31, "sex": "M"}
|
346 |
+
"""
|
347 |
+
|
348 |
+
fields: List[str]
|
349 |
+
to_field: str
|
350 |
+
use_query: bool = False
|
351 |
+
|
352 |
+
def process(
|
353 |
+
self, instance: Dict[str, Any], stream_name: Optional[str] = None
|
354 |
+
) -> Dict[str, Any]:
|
355 |
+
keylist = dict_get(instance, self.fields[0], use_dpath=self.use_query)
|
356 |
+
valuelist = dict_get(instance, self.fields[1], use_dpath=self.use_query)
|
357 |
+
|
358 |
+
output_dict = {}
|
359 |
+
for key, value in zip(keylist, valuelist):
|
360 |
+
output_dict[key] = value
|
361 |
+
|
362 |
+
instance[self.to_field] = output_dict
|
363 |
+
|
364 |
+
return instance
|